PSYCHOLOGICAL PREDICATES AND VERBALCOMPLEMENTATION IN ARABIC

Yasir Alotaibi, Muhammad Alzaidi, Maris CamilleriShaimaa ElSadek and Louisa Sadler

University of Essex

Proceedings of the LFG13 Conference

Miriam Butt and Tracy Holloway King (Editors)

2013

CSLI Publications

http://csli-publications.stanford.edu/

Abstract

The issue of verbal complementation patterns in the Arabic vernaculars isone which is relatively under-researched: this paper aims to make a small con-tribution in this area, focussing on essentially two issues (i) the syntax of so-called experiencer-object psychological predicates (EOPVs) (that is, predicatesin the frighten or please classes) and (ii) the syntax of aspectual or phasal pred-icates (that is, verbs such as begin and continue). We argue that the latter classof verbs are in fact raising verbs and go on to show that in some dialects theinteraction of EOPV and aspectual predicates permits a pattern reminiscent ofCopy Raising.

1 IntroductionThe issue of verbal complementation patterns in the Arabic vernaculars is one whichis relatively under-researched: this paper aims to make a small contribution in thisarea, focussing on essentially two issues (i) the syntax of so-called experiencer-objectpsychological predicates (EOPVs) (predicates in the frighten or please classes) and (ii)the syntax of aspectual or phasal predicates (that is, verbs such as begin and continue).Our work concentrates on the complementation patterns for these classes of verb inthree geographically diverse dialects, Hijazi Arabic (a Gulf dialect from the West ofSaudi Arabic, henceforth HA), Egyptian Cairene Arabic (henceforth ECA) and Mal-tese (henceforth MT). EOPVs are known to exhibit unusual properties crosslinguis-tically, and we will explore the extent to which this is true for Arabic and provideevidence that the experiencer really is a normal OBJ in this class of verbs. As for theaspectual verbs, we will argue that they are raising predicates in Arabic. We will thenshow that in some dialects, the interaction of EOPVs with aspectual verbs shows a pat-tern highly reminiscent of copy raising, although this in turn raises a number of openquestions about the correct approach to the analysis of such constructions. Through-out, our principal aim is not theory development but a relatively detailed descriptionof some under-studied verbal complementation patterns.

2 Psychological predicatesThe term psychological predicates refers to those classes of predicates with an argu-ment structure or thematic role grid involving an experiencer and a theme or stimulusargument (the content or object of the mental state). Verbs such as fear and Italiantemere ‘fear’, which map their arguments so that the experiencer is the SUBJ andthe theme or stimulus is the OBJ contrast sharply with verbs such as frighten or Ital-ian preoccupare ‘worry’, which exhibit the inverse mapping, with the experiencer asOBJ and the theme or stimulus as SUBJ. Furthermore this ‘inverted’ mapping occurswhether or not the stimulus is interpreted causally. A third class of verbs also exhibit-ing an apparently ‘inverse’ mapping include those which mark the experiencer with a

†A version of this work was presented at the 6th International Arabic Linguistics Symposium, inIfrane, Morocco. We thank participants at ALS and at LFG2013 and our reviewers for comments andfeedback. All remaining errors are our own.

preposition or dative case marker, for example Italian piacere ‘please’. These classesare illustrated with the examples from Belletti and Rizzi (1988, 291).1

(1) GianniGianni

temefears

questothis

(2) Questothis

preoccupaworries

GianniGianni

(3) Ato

GianniGianni

piacepleases

questothis

(4) Questothis

piacepleases

ato

GianniGianni

In common with other literature we use the term EOPV to refer to predicates inboth of these last two classes, that is, as a superordinate term for the frighten, preoc-cupare and the piacere classes (Belletti and Rizzi (1988)’s Classes 2 and 3). Class3 (piacere) predicates are always stative and class 2 predicates (frighten, preoccu-pare) are usually ambiguous between stative and eventive readings. A representativesample of the class of predicates is given in Table 1.

Meaning HA ECA MTanger yugdib, yuzaQil - jagèdabovercome yaglib, yusait

˙ir yiglib jegèleb

tease/annoy/bother yuzQig, yuqliq yigız, yidaayeP jdejjaqtire yatQib yetQib jgèejjahurt yagrah

˙yigrah

˙jwegga’

frighten yah¯awif yixawwef jbezza’

make happy/please yafarrih˙

yifarrah˙, yibsit jferraè

like yah˙ib yiQgib jogègob

make sad yah˙zin, yuQlim yizaQQal jnikket

enable yumakin yiQaddar -

Table 1: Experiencer Object Psych Predicates

Arabic verbal morphology is characterised by a system of forms (Arabic Pawzan(sg: wazn) or Hebrew binyanim (sg: binyan)) involving derivational morphologicalprocesses by which new verbal lexemes are formed. In the Western tradition, theseforms (or measures) are referred to by means of roman numerals, with the Ist form be-ing the basic underived lexeme. In Arabic we find some fear-frighten pairs expressedthrough a change in binyan, such as ECA xaaf ‘fear’ < SUBJ OBL> (Ist binyan)and xawwef ‘frighten’ <SUBJ, OBJ> (IInd binyan) or MT beza’ ‘fear’ (Ist binyan) -bezza’ ‘frighten’ (IInd binyan).2 In other cases, we find that the same verbal stem

1For Belletti and Rizzi (1988) verbs in the fear class have the experiencer as an external argument,while the remaining two classes lack an external argument and associate the experiencer argument withACC and DAT case respectively. For piacere the experiencer can ‘end up’ in preverbal subject positionwhile for the preoccupare class the theme is generated in canonical object position but ends up inpreverbal position, while the experiencer is sister of V′.

2These two EOPVs are IInd binyan derived verbs. Amongst the semantic shifts associated withthe use of the IInd binyan is causativisation, but it is also associated with a number of other semanticshifts. Note that EOPVs in the IInd binyan are not necessarily interpreted as agentive or causative.

simply permits both the experiencer-SUBJ mapping and the EOPV mapping.The following examples from ECA illustrate the class of EOPV predicates we are

concerned with. In (5) it is el-walad ‘the boys’ that controls verbal agreement asSUBJ of the matrix clause.3 Note however that, just as in other languages, an animateSUBJ stimulus does not necessarily entail an agentive or causative interpretation. In(6) the experiencer argument occurs as an OBJ incorporated pronominal -ha and theSUBJ is the inanimate theme/stimulus el-saraèa ‘frankness’.

(5) el-weladDEF-boys

bi-dayP-uBI-annoy.IMP.3-PL

el-banatDEF-girls

The boys annoy the girls. ECA

(6) b-t-Qegeb-haBI-3SGF-like.IMP-3SGF.ACC

el-sarah˙a

DEF-franknessShe likes frankness. ECA

The stimulus argument may be propositional, with default 3SGM agreement onthe verb as in (7) and (8). Although we will not pursue this matter here, exampleswith a propositional stimulus argument raise the interesting question of whether thepropositional argument is the subject or whether these are essentially extrapositionstructures with an expletive subject and an XCOMP or COMP argument. We note thatit is possible for the experiencer argument to correspond to the SUBJ of the embeddedproposition, as in (9)-(10).4

(7) bi-dayeP-niBI-annoy.IMP.3SGM-1SG.ACC

PinCOMP

el-sotDEF-sound

yi-kun3-be.IMP.SGM

Qalyloud.SGM

It annoys me that the sound is loud. ECA

(8) ya-Qgib-ha3-like.IMP.3SGM-3SGF.ACC

PinCOMP

MuèammadMuhammad

d˙al

remain.PV.3SGM

sakitquiet.ACT.PRT.SGM

It pleases her that Muhammad remained quiet. HA

In the dialects, the morphological system of binyanim or Pawzan is basically a system of templaticphonological structures with appended affixes. The IInd binyan involves gemination of the verb’s secondradical.

3The gloss BI marks a particular indicative realis form of the imperfective verb. This matter isdiscussed in some detail below. Other glosses are standard.

4Further interesting issues are raised by examples such as (i), in which the psych verb agrees withMarija, suggesting that this NP is the SUBJ. The question is whether this subject is raised from theembedded predication, which we would then expect to be an XCOMP, or whether on the other hand thematrix SUBJ is thematic and the clause has the status of an adjunct. We leave this matter for futureresearch.

(i) MarijaMary

d-dejjaq-hom3-annoys.IMP.SGF-3PL.ACC

to-èrog3-goes.out.IMP.SGF

weèid-haalone-3SGF.ACC

Mary annoys them going out on her own. MT

(9) ya-Qgib-ha3-pleases.IMP.3SGM-3SGF.ACC

SarahSarah

ta-h¯rog

3SGF-go.out.IMP

li-wah˙da-ha

for-alone-3SGF.ACC

It pleases Sarah to go out alone./Going out alone pleases Sarah. HA

(10) JienI

j-o-gègob-ni3-FRM.VWL-pleases.IMP.SGM-1SG.ACC

n-o-èrog1-FRM.VWL-go.out.IMP.SG

waèd-ialone-1SG.ACC

It pleases me to go out alone.Going out alone pleases me. MT

Although the stimulus or theme does correspond to the SUBJ with verbs in thisclass, it is worth noting that it does not always appear in the canonical SUBJ positionin terms of word order in Maltese. While both SVO and VSO (and indeed other orders)are quite freely available in ECA and HA, Maltese is a predominantly SVO language.However there appears to be a marked preference for the SUBJ stimulus of these verbsto follow the verb, as in example (14). Note that such a postverbal SUBJ is really partof the matrix sentence, and not in a clause-external dislocated position (althoughthe language makes extensive use of such dislocation structures involving externaltopics). Nonetheless, this (postverbal position) is not an invariant requirement asexamples such as (11) have the experiencer in postverbal position and the stimulus ortheme preverbally.

(11) Xisome

kliemword.SGM

liCOMP

nt-qalPASS-said.PV.3SGM

dejjaqannoyed.PV.3SGM

lin-niesACC.DEF-peopleSome words that were said annoyed the people. MT

For the majority of the EOPV verbs we have investigated, the experiencer is codedas an OBJ. This can be seen in examples such as (6)-(10), where the pronominal ex-periencer is coded by means of an OBJ inflection on the verbal element. The contem-porary Arabic dialects do not exhibit case marking, but the corresponding nominalswould be marked with ACC case in Modern Standard Arabic (MSA). For some verbs,however, the experiencer is either an OBL or marked by the dative (i.e. goal or re-cipient) marker/preposition li-. Note that in MT, pronominal li-marked arguments arealso expressed inflectionally as part of the verbal form.5

5Elsewhere we have argued that recipient/goal li-marked arguments in Maltese are actually instancesof the grammatical function OBJθ rather than OBL (Sadler and Camilleri, 2013). For some discussion ofthe possibility that this might also be the case in ECA see Camilleri et al. (2013). We will have nothingfurther to say here on this question.In relation to (14), a reviewer questions our assumption that this is a psych predicate in this context,suggesting that li may mark ’a recipient or goal’ here, a function which it certainly has in other contexts.While we believe that it does mark an experiencer in this example, detailed discussion of this examplewould take us too far afield. A further example of a psych verb with a DAT-marked experiencers is

(12) kabascompress.PV.3SGM

Qale-haon-3SGF.ACC

al-noumDEF-sleep

Sleep overcame her. ECA

(13) s˙ar

become.PV.3SGM

yu-sait˙ir

3-overcome.IMP.SGM

Qalaon

MuhammadMuhammad

al-nawmDEF-sleepiness

Muhammad is overcome by sleepiness. HA

(14) Naqas-l-ireduce/lack.PV.3SGM-DAT-1SG

d-dawlDEF-light

Lit: The light reduced to meI am experiencing increased blindness. MT

Before leaving the question of pronominal experiencers, it is worth noting in pass-ing that while these are expressed as verbal affixes in neutral discourse conditions, itis possible in Maltese to use a full pronoun in cases of contrastive focus such as (15).

(15) LILHA

hergèogobplease.PV.3SGM

il-ktieb,DEF-book.SGM

uCONJ

mhuxNEG

lili.me

It was she who liked the book and not me. MT

In a recent book on experiencers, Landau (2010) proposes that many of the un-usual properties that experiencer objects exhibit crosslinguistically follow if the ex-periencer arguments of non-agentive (readings of) psych verbs are not OBJ but aretaken to be underlyingly obliques, that is, objects of an abstract locative preposition,as mental locations. In particular, he argues for this position in Modern Hebrew, arelated Semitic language. However, data from the Arabic dialects we are concernedwith does not appear to support the extension of this abstract analysis to Arabic. Aswe have already noted, evidence from the surface forms supports the view that theexperiencer is straightforwardly an OBJ; in particular, pronominal experiencers areverbal inflections. Objects of prepositions are expressed as prepositional inflections(and these inflectional paradigms are not identical, at least in the form used to real-ize the 1SG set of values).6 A further piece of robust evidence is the fact that theexperiencer argument may be the SUBJ under passivisation of predicates in this class

appella ’appeal’, as in (i):

(i) Appella-t-l-iappealed.PV-3SGF-DAT-1SG

ferma.lot

dil-esperjenzaDEM.SGF.DEF-experience.SGF

This experience appealed to me a lot.

6It must of course be acknowledged that the use of ACC morphology does not provide irrefutableproof of GF status, especially given that in Maltese there is a set of defective verbs (the ‘pseudo-verbs’) (Peterson, 2009) which take ACC pronominal markers in what is probably a SUBJ functione.g. donn-ok, donn-hom, donn-ha ’appear’/’seem’, il-ek, il-u, il-na ’long.time’ and qis-ni, qis-kom, qis-ha ’as.though/look like/appear’. Nonetheless the general point is clear - the morphological evidence ismost consistent with the OBJ rather than the OBL analysis of the experiencer arguments.

(despite the fact that Belletti and Rizzi (1988, 309) claim that experiencer object verbscannot be passivised). (16)-(17) illustrates this with a HA active-passive pair involv-ing a shift from the Ist to the VIIth binyan and (18)-(19) an active-passive pair in MT

involving a shift from the IInd to the Vth binyan. (In (18) and other subsequent ex-amples, the parenthesised NPs indicate typical positions for the NP, which may alsobe dropped.)7

(16) al-filmDEF-film

ya-fgaQ-ha3-frighten.IMP.SGM-3SGF.ACC

The film frightens her. HA

(17) n-fagaQ-atPASS-frighten.PV-3SGF

minnfrom

al-filmDEF-film

She was frightened by the film. HA

(18) (LilACC

Mario)Mario

t-bezzgè-u3-make.fear.IMP.SGF-3SGM.ACC

l-mewtDEF-death.SGF

(lilACC

Mario)Mario

Death frightens Mario. MT

(19) MarioMario

dejjemalways

t-bezza’PASS-cause.fear.PV.3SGM

mill-mewtfrom.DEF-death.SGF

Mario was always frightened by death. MT

One special property of the experiencer which Landau (2010, 5) interprets asfavouring an OBL analysis concerns the distribution of resumptive pronouns (RPs).This also holds in Maltese, and for this reason we mention the relevant data here, al-though it is not clear to us that any analytic consequences in terms of GF follow fromthis observation. Landau notes that in Hebrew, while a RP is typically optional in OBJ

position within relative clauses, a RP encoding an experiencer object is obligatorilypresent. In this respect, the experiencer appears to behave more like an oblique, since

7Passivisation in Classical Arabic and MSA involves the use of specific vowel patterns but this strat-egy is largely (although not entirely) absent in the contemporary vernaculars, where derivational pro-cesses in the binyanim system are generally used for verbal diathesis alternations. In Classical Arabicand MSA these same binyanim fulfill other broadly intransitivising functions. The fact that these formsyield passives in the dialects is well established in the literature (see Holes (2004, 135-138) and Abdel-Massih (1979/2011, 195)). Further evidence can be provided from MT, which also has a syntacticpassive formed from the use of gie ’come’ and the passive participle. The following pair shows thepromotion of the experiencer to SUBJ of the syntactic passive.

(i) a. J-bezzagè-ni3-make.fear.IMP.SGM-1SG.ACC

l-fattDEF-fact.SGM

liCOMP

èaPROSP

m-mut-u1-die.IMP-PL

The fact that we will die frightens me.

b. Gej-tcome.PV-1SG

im-bezza’PASS.PRT-fear.SGM

mill-fattfrom.DEF-fact

liCOMP

èaPROSP

m-mut-u1-die.IMP-PL

I was frightened from the fact that we will die.

obliques involve an obligatory RP. A similar pattern is found in MT: normally an ob-ject relative clause with a definite head noun would involve a gap, but the experiencerOBJ of an EOPV requires an obligatory RP. We give the data in (20)-(21) but as noted,it is not clear what to make of this observation.

(20) Kellim-tspoke.PV-1SG

lit-tifelACC.DEF-boy

liCOMP

wegggè-et-umake.hurt.PV-3SGF-3SGM.ACC

ras-uhead-3SGM.ACC

//

gèajr-u-htease/call.out.names.PV.3-PL-3SGM.ACC

xisome

subienboys

ilbieraè

yesterdayI spoke to the boy whose head was hurting yesterday/who some children teasedyesterday. MT

(21) Kellim-tspoke.PV-1SG

lit-tifelACC.DEF-boy

liCOMP

ra-t-*usaw.PV-3SGF-*3SGM.ACC

omm-imother-1SG.ACC

ilbieraè

yesterdayI spoke to the boy who my mother saw yesterday. MT

When we turn to data involving binding and scope, the pattern which emergesis one in which the experiencer OBJ patterns alongside the OBJ of other verbs, con-trary to the special behaviour which is reported for (non-agentive) EOPVs in otherlanguages. Here we merely summarise the situation as it appears in the data we haveexplored: very little is clear about the syntactic hierarchical and linear conditions onbinding and scope in Arabic in general. We can conclude, however, that the accessi-bility relations for experiencer objects are little or no different from those for standardtransitive objects, supporting the view that such arguments are indeed OBJs. Considerfirst so-called backward binding, where an EO but not a ‘normal’ object can binda reflexive within the subject (compare Pictures of herself pleased Mary with *Pic-tures of herself hit Mary). In the dialects we consider, on the other hand, there is nodifference in this regard between EOPV objects and other objects, as shown in the MT

examples in (22) and ECA (23).

(22) a. Dal-kliemDEM.SGM.DEF-words.SGM

dwar-uabout-3SG.ACC

nnifs-uibreath-3SG.ACC

dejjaqbother.PV.3SGM

lilACC

PawluiPaul

//

dejjq-uibother.PV.3SGM-3SGM.ACC,

’lACC

PawluPaul

These words about himself bothered Paul. MT

b. èolmadream.SGF

dwar-uabout-3SG.ACC

nnifs-uibreath-3SG.ACC

qajjm-etmake.wake.PV-3SGF

’lACC

PawluiPaul

b’èasdawith.shock

A dream about himself woke up Paul all of a sudden. MT

(23) QaliAli

dayP-uannoy.PV.3SGM-3SGM.ACC

el-kalamDEF-talk

Qanabout

nafs-uself-3SGM.ACC

Ali was annoyed by the talk about himself. ECA

A second data constrast is argued to arise for English in terms of weak crossover.EOPV (but not other objects) allow crossover violations, inverse variable binding ofthe reflexive within the subject by the experiencer argument. However our initial in-vestigations suggest that EOPV pattern just like other transitive verbs in not permittingweak crossover violations, as shown in (25)-(27).

(24) a. Hisi promotion pleases everyoneiHisi health worries every patienti

b. *Hisi father hit everyonei∗Hisi father killed everyonei

(25) a. Puslub-uhibehaviour-3SGM.ACC

bi-gızBI-annoy.IMP.3SGM

kolevery/each

wah˙edj

oneHis behaviour annoys everyone/each one. ECA

b. Pumm-uhimother-3SGM.ACC

gawb-etanswer.PV-3SGF

kolevery/each

wah˙edj

oneHis mother answered everyone/each one. ECA

(26) a. soluk-uhibehaviour-3SGM.ACC

ya-zQig3-annoy.IMP.3SGM

kulevery

wah˙idj

oneHis behaviour annoys everyone. HA

b. Pumm-uhimother-3SGM.ACC

gawab-atanswer.PV-3SGF

kulevery

wah˙idj

oneHis mother answered everyone. HA

(27) a. Imgiebt-uibehaviour.SGF-3SGM.ACC

gèadb-etangered.PV-3SGF

’lACC

kulèaddjeveryone

His behaviour angered everyone.

b. Omm-uimother-3SGM.ACC

wiegb-etanswered.PV-3SGF

lilACC

kulèaddjeveryone

His mother answered everyone. MT

A third issue concerns anti-local binding. It is claimed that non-agentive EOPVsdo not allow local binding of an anaphor in direct object position by the causer subject(see (28)). But here too, Arabic EOPVs appear to permit local binding in a manneranalogous to other verbs. Contrary to Landau (2010)’s ‘universal’ claim that this is

not possible (on non-agentive readings), examples such as (29) and (30) have non-agentive readings and involve local binding.8

(28) a. ??John amuses/disgusts/horrifies/irritates himself.

b. John killed/hurt himself.

(29) Muh˙ammadi

Muhammadbi-yi-tQibBI-3-tire.IMP.SGM

nafs-uiself-3SGM.ACC

Muhammad tires himself. ECA

(30) In-dejjq-ui1-bother.IMP-PL

lilACC

xulxinieach.other

//

lilnaus

nfus-naibreath-1PL.ACC

xisome

kultanttime

We bother each other/ourselves sometimes. MT

Our investigation of the syntactic properties of these verbs, in which the experi-encer argument maps to a lower function than SUBJ leads us to conclude that there isgood evidence that the experiencer is a bona fide OBJ for verbs in this class. In par-ticular, it appears to lack many of the special properties often ascribed to EOs. In thefollowing section, we turn to a completely different set of verbs which embed verbalcomplements, before turning in section 4 to the interaction of these two classes ofpredicates.

3 Aspectual or Phasal verbsBy aspectual or phasal verbs we refer to a cass of predicates which take a verbalcomplement and which denote the inception, duration, continuation or termination(and so on) of an event or state. Such verbs are typically either PRED-less auxiliaries,or (more often) control or raising predicates. A representative sample of verbs inthis class for the dialects we discuss is given in Table 2; (31) and (32) exemplify thestructure.

Meaning HA ECA MTbegin bada/qam badaP beda/qamremain qaQid/d

˙al PaQad/fedel baqa’/fadal

finish/achieve bat˙t˙al/lih

˙iq battal/leh

˙eP laèaq

repeat - regeQ rega’(be)near/almost qarrab Parrab qorobbecome s

˙ar baPa sar

Table 2: Aspectual/Phasal Verbs

8A fourth observation is that EOPV but not standard transitives permit both scopings of SUBJ andOBJ in cases such as (8) (Kim and Larson, 1989).

(i) a. What worried everyone? what> ∀, ∀ > what

b. Who hit everything on purpose? who > ∀, *∀ > what

We have not yet investigated this for the dialects.

(31) Bedabegin.PV.3SGM

j-i-gbor3-FRM.VWL-gather.IMP.SGM

l-iltiemaDEF-orphans

He started gathering the orphans. MT

(32) el-weladDEF-boy.PL

badaP-ustart.PV-3PL

ya-kl-u3-eat.IMP-PL

The boys started to eat. ECA

The most salient properties of this class of verbs include the fact that they are timereference dependent. Al-Aqarbeh (2011) provides extensive discussion of the com-plementation patterns of Jordanian Arabic (JA), documenting this property (amongstothers) for JA ballash ‘begin’, see (33), and other verbs in this class.

(33) QaliAli

ballashbegin.PV.3SGM

yi-ktub3-write.IMP.SGM

i-risalihthe-letter

Ali began to write the letter. (Al-Aqarbeh:128) JA

Further salient properties are that (i) they take verbal (or nominalised verbal) com-plements, (ii) typically nothing intervenes betwen the aspectual verb and its verbalcomplement, (iii) that generally, there is no embedded complementiser, (iv) the as-pectual verb and the embedded verb have the same SUBJ, which is not expressed asan NP in the lower clause, and (v) the embedded verb shows subject agreement andis a morphologically finite form.

Arabic does not have an infinitival verb form, although it does have a nominal(verb-noun) form, the masd

˙ar, and participle forms. Morphologically, the basic con-

trast is between the perfective and the imperfective stem. In Classical Arabic (andMSA) the imperfective stem is used to form the imperfective indicative, the futureand two ‘moods’ — the jussive and the subjunctive in the Western tradition. Thesemoods are essentially dependent verb forms used in a variety of contexts.9 The di-alects which we discuss here all display a basic contrast between perfective, imper-fective and future (the form of the latter involving a prefix added to the imperfectiveforms). In addition, ECA (and other dialects such as JA) distinguish between an imper-fective form with a bi- prefix, which is used in most indicative declarative contextsand seems to be essentially a realis form, and a ‘bare’ imperfective form, which isused in many modal and embedded contexts, and may be thought of as an irrealisform. We simply gloss the former form as BI. Formally, the distinction made by theClassical Arabic system of moods built on the imperfective verb form does not existin HA (as far as we are aware) and MT. The verbal complements to the class of as-pectual verbs across all three dialects are usually the imperfective forms (and in ECA

usually bare imperfective forms), but perfective and future forms are not completelyexcluded.

9Compound tenses are formed using combinations of imperfective and perfective verbs with perfec-tive and imperfective forms of kan ‘be’. The fact that the basic perfective and imperfective forms can beused to relate speech time to reference time and to relate reference time to event time provides significantevidence that morphological forms must be sharply distinguished from their (multiple) interpretations.See Fassi Fehri (2012) for some discussion of the Arabic tense and aspect system.

Evidence that the aspectual verbs have PRED values and hence head their ownf-structures (rather than forming complex predicates with the lexical verb or intro-ducing only featural information under a co-head analysis) comes from their interac-tion with modifiers and with negation. Consider the meaning distinction which arisesbetween (34a) and (34b), and similarly between (35a)-(35b) and (36a)-(36b), accord-ing to whether the aspectual verb or the embedded verb is negated. Example (37)shows that modification may target the predicates separately (and this is true also ofthe other dialects).

(34) a. el-waladDEF-boy

ma-badaP-sNEG-start.PV.3SGM.NEG

ya-kul3-eat.IMP.SGM

The boy didn’t start to eat. ECA

b. el-waladDEF-boy

badaP

start.PV.3SGM

ma-ya-kul-sNEG-3-eat.IMP.SGM.NEG

The boy started to not eat. ECA

(35) a. maNEG

d˙al

remain.PV.3SGM

Pa-kallim-ak1SG-speak.IMP-2SG.ACC

kulevery

yawm.day

I will not remain/stay talking to you every day. HA

b. qaQidstay.PV.3SGM

PaliAli

maNEG

ya-kallim-niIMP.3-speak-1SG.ACC

lifor

modatperiod

PisbuQ.week

Ali stayed not talking to me for one week. HA

(36) a. MaNEG

n-o-qgèod-x1-FRM.VWL-stay.IMP.SG-NEG

in-kellm-ek1-speak.IMP.SG-2SG.ACC

darb’oèr-aonce.SGF.another-SGFI won’t stay speaking to you next time.I will not endure speaking to you next time. MT

b. Qagèadstayed.PV.3SGM

maNEG

j-kellim-ni-x3-speak.IMP.SGM-1SG.ACC-NEG

gèall-gimgèafor.DEF-week.SGF

sèiè-acomplete-SGFHe stayed not talking to me for a whole week. MT

(37) a. Jekkif

j-i-bde-w3-FRM.VWL-start.IMP-PL

kullevery

darbaonce

j-morr-u3-go.IMP-PL

l-gèadaDEF-tomorrow

If they start all the time going the next day ... MT

b. Jekkif

j-i-bqa’3-FRM.VWL-stay.IMP.SGM

l-èinDEF-time.SGM

koll-uall-3SGM.ACC

j-iekol3-eat.IMP.SGM

èafnaa.lot

If he remains all the time eating a lot.... MT

If we are correct in arguing that the aspectual verb has a PRED value, the ver-bal complement (with controlled subject) could in principle be a case of control orraising. We believe that the evidence favours the conclusion that this is in fact raising.

First, at least some verbs in this class, but not all, allow a variant structure inwhich the aspectual verb takes default 3SGM agreement (and no NP subject) andthe embedded verbal complement may contain its own independent subject. (35a)above illustrates that the HA d

˙al ‘remain’ permits this: others include MT baqa’/fadal

’remain’, sar ’become’ and laèaq ’achieve’.10 If these are raising verbs, then bothvariants have the same set of thematic or semantic arguments. But under a controlanalysis, the two variants of the verb do not share a single thematic/semantic argu-ment frame underlying the surface valency patterns. (38) shows that both patternsoccur with MT sar ‘become’, in contrast with HA s

˙ar ‘become’, which does not

permit the 3SGM default/impersonal pattern. In (40) both the aspectual verb andthe embedded verb show 3SGF agreement with the subject ‘she’, in contrast to (41),which shows default agreement on the aspectualiser.

(38) Sar/sar-etbecome.3PV.SGM/become.PV-3SGF

maNEG

t-i-swa3-FRM.VWL-cost.IMPV.SGF

xejnnothing

il-èajjaDEF-life.SGF

It became such that life costs nothing, i.e. is futile. MT

(39) s˙ar-at/*s

˙ar

become.PV-3SF/*become.PV.3SGM

maNEG

ti-swa3SGF-cost.IMP

h˙agah

thingal-h

˙ayah.

DEF-lifeIt becomes such that life costs nothing. HA

(40) QedPROG

t-i-bqa’3-FRM.VWL-remain.SGF

t-webbes3-make.hard.IMP.SGF

ras-hahead-3SGF.ACC

Lit: She is continuing to harden her head.She keeps being hard-headed. MT

(41) Baqa’remain.PV.3SGM

n-a-gèmel1-FRM.VWL-do.SG

danDEM.SGM

koll-uall-3SGM.ACC

Lit: It remains I do all this.I still have to do all this. MT

For MT in particular, we find structures with multiple embedded impersonal (de-fault 3SGM) verb forms, with the thematic argument expressed by a DAT-markedpronoun, suggesting that these verbs impose no selectional restrictions on their SUBJ.

10There is very little literature on the topic of subjectless, expletive subject or impersonal construc-tions in Arabic. However see Firanescu (2010) for discussion of a similar pattern with a subset ofaspectual verbs in Syrian Arabic.

(42) HaPROSP

j-e-rga’3-FRM.VWL-repeat.IMP.SGM

j-i-bda3-FRM.VWL-start.IMP.SGM

j-kol-l-i3-be.DAT-1SG

marawoman

t-gèin-ni3-help.IMP.SGF-1SG.ACC

fid-darin.DEF-house

Lit: It will repeat start it be to-me a woman helps me in the houseI will again start having a woman helping me in the house MT

Second, aspectual predicates (irrespective of whether they permit impersonal con-structions such as those illustrated in (41) above) do not appear to impose any selec-tional restrictions on their subjects, permitting human, inanimate and idiom chunks(preserving idiomatic meaning). Inanimate subjects are shown in (43) and (44) andan idiom chunk in (45).

(43) al-bardDEF-cold

badastart.PV.3SGM

ya-gi.3-come.IMPV.SGM

It started being cold. HA

(44) Baqgè-etremain.PV-3SGF

t-a-gèmel/niezla3-FRM.VWL-do.IMP.SGF/PROG.PRT.falling.SGF

x-xitaDEF-rain.SGF

The rain continued falling/It continued to rain. MT

(45) AllaGod

skontaccording

il-muntanjaDEF-mountain

(j-i-bqa’)(3-FRM.VWL-remain.IMP.SGM)

j-ti-ha3-give.IMP.SGM-3SGF.ACC

s-silgDEF-ice/snow

Lit: God keeps giving snow according to the mountain.God will never give you more than you can handle. MT

Third, the passivisation test supports the conclusion that these verbs are instancesof raising: the version with an active embedding in (46) and the corresponding passiveembedding in (47) are equivalent in meaning in the sense that they describe the sameevent, as is expected with raising verbs but not with control.11

(46) el-waladDEF-boy

badaP

start.PV.3SGM

ya-kul3-eat.IMP.SGM

el-PaklDEF-food

The boy started to eat the food. ECA

11Note that the aspectual verb ‘start’ in (47) is given here in the VIIIth binyan, but this is not itself apassive: it is the embedded verb which is passive. It would also be possible to use an (underived) Istbinyan form here, although the resultant sentence is less natural.

(i) el-PaklDEF-food

badaP

start.PV.3SGM

yi-t-akel3-PASS-eat.IMP.SGM

The food started to be eaten. ECA

(47) el-PaklDEF-food

Pibtadastart.PV.3SGM

yi-t-akel3-PASS-eat.IMP.SGM

The food started to be eaten. ECA

(48) a. badastart.PV.3SGM

ya-gamiQ3-gather.IMP.SGM

al-mah˙s˙ul

DEF-harvestHe started gathering the harvest. HA

b. al-mah˙s˙ul

DEF-harvestbadastart.PV.3SGM

ya-n-gimQ

3-PASS-gather.IMP.SGM

The harvest started being gathered. HA

(49) a. Bedabegin.PV.3SGM

j-i-gbor3-FRM.VWL-gather.IMP.SGM

l-iltiemaDEF-orphans

He started gathering (together) the orphans. MT

b. L-iltiemaDEF-orphans

bde-wbegin.PV.3-PL

j-i-n-gabr-u3-FRM.VWL-PASS-gather.IMP-PL

The orphans started being gathered (together). MT

This section has considered the behaviour of a class of verbal predicates, the as-pectual or phasal verbs, all of which take same subject verbal complements, whilea small number of them also permit a construction with an impersonal or expletivesubject. These temporally dependent complements are usually, but not always, in theimperfective form. In ECA, which distinguishes a clearly tensed realis form of theimperfective (using the verbal prefix bi-) from a dependent form of the imperfective(used inter alia in modal contexts), it is the dependent form of the imperfective whichis used. Standard tests for distinguishing cases of raising from control support theconclusion that verbs in this class are raising verbs. We suggest that this is indeedthe case. There is no reason in principle to reject a raising analysis on the basis ofthe embedded verbal morphology. First, there is substantial evidence in the litera-ture that languages including Greek, Romanian, Bulgarian, Nguni, Shona, Kikuyuand Kirundi have finite raising or hyperaising. Second, we must clearly distinguishthe use of particular surface forms (that is, finiteness as a inflectional property ofverbs) from grammatical content (that is, FINITENESS as a property of a clause indiscourse); see Sells (2007) for discussion of this point. The use of morphologicallyfinite verb forms in Arabic does not then necessarily entail that these verbal comple-ments are syntactically FINITE, and certainly does not rule out a raising analysis usingfunctional control, even if it should turn out that they are in fact syntactically finite.Arka (2000) suggests an f-control analysis for raising out of finite complements inIndonesian, while on the other hand Asudeh (2005, 495) proposes that all cases offinite control should be analysed as obligatory a-control. On the basis of the obser-vations made in this section, we suggest that these aspectual verbs are indeed raisingpredicates.

4 EOPV complements to Aspectual PredicatesIn section 2 we argued that a certain class of psychological predicates in the Arabicvernaculars ECA, HA and MT are EOPVs: the stimulus or theme argument is SUBJ andthe experiencer argument is usually OBJ, sometimes OBJθ or OBL. In section 3 weargued that a class of aspectual or phasal verbs in these dialects are subject to subjectraising verbs (with a small number of verbs in this class also permitting a non-raisedconstruction, with default 3SGM verbal morphology). If these observations are alongthe right track, it is expected that the EOPV non-experiencer argument will raise inthe complement of an aspectual verb. This is shown in (50)-(52).

(50) el-weladDEF-boys

badaP-ustart.PV-3PL

yi-dayP-u3-annoy.IMP-PL

monaMona

The boys started to annoy Mona. ECA

(51) bad-ustarted.PV.3-PL

al-Pawladthe-boys

yu-d˙ayiq-u-n

3-annoy.IMP-PL-IND

monaMona

The boys started to annoy Mona. HA

(52) Is-subienDEF-boys

bde-wstarted.PV.3-PL

i-dejq-u3-bother.IMP-PL

’l-bnietACC.DEF-girls

The boys started to bother the girls. MT

Abstracting away from a number of details (for example simply using the glosslabels as VFORM:VALUE pairs), and assuming for the purposes of illustration anXCOMP f-control analysis rather than a COMP a-control analysis, the structure of suchexamples would be as in (53), and the lexical entry for a verb form such as that in(50) would include the information in (54).

(53)

PRED ‘BEGIN< XCOMP> SUBJ’VFORM PVSUBJ

XCOMP

VFORM IMP

PRED ‘ANNOY<SUBJ OBJ >’

SUBJ

PRED ‘BOYSNUM PL

PERS 3DEF +

OBJ

PRED ‘MONA’NUM SG

PERS 3

(54) badaP-u:

(↑ PRED) = ‘BEGIN< XCOMP> SUBJ’(↑ XCOMP VFORM) = IMP

However, the intriguing fact is that a further possibility is found robustly in HA

and ECA, but not in MT. In these cases, the EO in the embedded clause is apparentlyallowed as SUBJ of the matrix aspectual predicate, with the EO being a pronominalform that is co-referent with the matrix subject. Examples in (55)-(58) illustrate thispattern with various word orders: note that the aspectual verb agrees with Mona (theexperiencer of the embedded predication) and the psych verb agrees with the boys(the stimulus).

(55) badaP-etstart.PV-3SGF

monaMona

yi-dayP-u-ha3-annoy.IMP-PL-3SGF.ACC

el-weladthe-boys

Mona started to be annoyed by the boys. ECA (VSVS)

(56) monaMona

badaP-etstart.PV-3SGF

yi-dayP-u-ha3-annoy.IMP-PL-3SGF.ACC

el-weladthe-boys

Mona started to be annoyed by the boys. ECA (SVVS)

(57) monaMona

bad-atstart.PV-3SGF

ya-d˙ayiq-un-aha

3-annoy.IMP-PL-3SGF.ACC

al-Pawladthe-boys

The boys started to annoy Mona. HA (SVVS)

(58) monaMona

bad-atstart.PV-3SGF

al-Pawladthe-boys

ya-d˙ayiq-un-aha

3-annoy.IMP-PL-3SGF.ACC

The boys started to annoy Mona. HA (SVSV)

Although we have not (yet) made any systematic investigation of written MSA

sources, and we know of no literature on MSA (or indeed on any of the dialects) whichdiscusses the possibility of this unusual pattern of apparent raising, the followingexample, taken from Haddad (2012, 73), appears to illustrate a similar phenomenonin MSA. In the second conjunct of (59) (wa-badPa-at ya-glib-u-ha l-sawad-u l-kah

˙il)

the aspectual verb agrees with what is also the SGF object of the psych verb ya-glib-u ‘overcome’ while the psych verb agrees with its SGM subject l-sawwad-u l-kah

˙il

‘pitch blackness’. Note that here too, the raised subject also occurs as an OBJ affix onthe embedded predicate.

(59) Sawwad-ublacken.PV.3-MPL

l-malamih˙-a

DEF-features-ACC

l-Parabiyat-aDEF-Arab-ACC

wa-badPa-atand-started.PV-3SGF

ya-glib-u-ha3-overcome.IMP-IND-3SGF.ACC

l-sawad-uDEF-blackness-NOM

l-kah˙il

DEF-pitchThey tarnished the Arab face, and it started to look pitch black. MSA

The existence of this construction, which to our knowledge has not been discussedin the literature on (varieties of) Arabic, raises many interesting analytic questions,which we cannot address in full here. In particular, it has a number of characteris-tics in common with Copy Raising (CR), and these commonalities suggest a possible

analysis.12 In recent work, Asudeh (2012) and Asudeh and Toivonen (2012) distin-guish English copy raising, illustrated in (60), from a complementation pattern foundwith perceptual resemblance verbs (such as look, sound,...), illustrated in (61).

(60) Chris seemed like he enjoyed the marathon.John seems like the judges ruled that he defeated Mary.John seems like Mary defeated him.

(61) John looked/sounded/smelled like Bill had served asparagus.

Copy raising with verbs such as seem, appear has the following characteristics:(i) a pronominal copy of the raised subject is found in the complement of the copyraising verb (according to Asudeh and Toivonen (2012) the copy is obligatory fornearly all speakers of English); (ii) the copy raised subject must be interpreted asa perceptual source (PSOURCE). Note that PSOURCE is not a thematic argument ofthe copy raising verb, but is an entailed participant in the state that the verb denotes(Asudeh and Toivonen, 2012, 334). On the other hand, Asudeh and Toivonen (2012)argue that this argument is thematic in the case of the perceptual resemblance verbs(see inter alia Landau (2011) for a different view on copy raising verbs and the notionof thematic argument).

A striking aspect of the Arabic construction we focus on here relates to this keynotion of a PSOURCE, which seems to be applicable to the circumstances in whichthese ‘raised object’ constructions arise. Asudeh and Toivonen (2012) note for En-glish and Swedish that “a copy raising subject is interpreted as the PSOURCE - thesource of perception - and ascribing the role of PSOURCE to the subject is infelicitousif the individual in question is not perceivable as the course of the report.” This alsoappears to hold for the distribution of this construction in Arabic. Examples such as(55)-(58) above are felicitious (roughly) when the state of affairs can be verified byinspection of Mona: that is, Mona is a perceptual source. This is naturally often thecase when the embedded predication is a psych verb. Although we are at an earlystage of investigating this pattern for other classes of verbs, we have found that ex-amples of this ‘raised object’ construction such as (62) and (63) are acceptable underparticular conditions, for example when inspection of the state of the car leads one toconclude that Ali has started driving it. This is perhaps suggestive of a connection tothe PSOURCE factor which is at work in English and Swedish copy raising, althoughthese remarks are necessarily highly speculative at this point.

(62) el-QarabeyyaDEF-car(SGF)

badaP-etstarted.PV-3SGF

yi-suP-ha3-drive.IMP.SGM-3SGF.ACC

QaliAli

The car started to be driven by Ali. ECA

12In addition to our discussion here, Sadler (2013) provides for a preliminary exploration of how theapproach of Asudeh (2012) and Asudeh and Toivonen (2012) might be extended to the Arabic data,although many questions remain unexplored.

(63) badaP-etstarted.PV-3SGF

el-QarabeyyaDEF-car(SGF)

QaliAli

yi-suP-ha3-drive.IMP.SGM-3SGF.ACC

The car started to be driven by Ali. ECA

In MT we do not find this construction at all with the aspectual/phasal verbs.However, it does seem to occur with verbs such as seem, happen. In (64) the matrixverb agrees with the experiencer of the embedded predication in the raised version,and may alternatively show default 3SGM agreement, with Marija, (if present), left orright dislocated from the embedded clause containing the OBJ pronominal. A furtherexample showing putative ‘object raising’ is given in (65).13

(64) (Marija)Mary

t-i-dher/j-i-dher3-FRM.VWL-appears.IMP.SGF/3-FRM.VWL-appears.IMP.SGM

(li)COMP

j-o-gègob-ha3-FRM.VWL-pleases.IMP.SGM-3SGF.ACC

dal-ktiebDEM.SGM.DEF-book.SGM

(Marija)Mary

Mary, it appears that the book pleases her.Mary appears such that this book pleases her. MT

(65) Seèl-u/seèelhappen.PV.3-PL/happen.PV.3SGM

qabad-homcatch.PV.3SGM-3PL.ACC

n-ngèas.SGM

DEF-sleepinessIt appears that they were overcome by sleepiness.They happened to be overcome by sleepiness. MT

A characteristic of CR (see the examples in (60)) is that the copy can be embeddedat some distance within the complement. We can show this to be the case with thepseudo-verbs of appearance such as donn- or qis-. These so-called pseudo-verbs inMT are aspectually defective predicates which use ACC bound pronominal forms tocode what is in fact (we believe) their SUBJ argument. The examples in (66) and (67)show that agreement on the matrix (pseudo-verb) predicate may be controlled by ei-ther argument of an embedded EOPV: that is, alongside the expected pattern in whichthe theme or stimulus is raised (donn-kom and qis-u, respectively), an alternative isgrammatical in which the experiencer controls agreement on the pseudo-verb. Thecrucial example is now (68): here an additional level of embedding is inserted intothe complement of the pseudo-verb by adding an aspectual predicate, and we see thatthe pseudo-verb may still agree with the (experiencer) object of the embedded psychpredication, showing the long distance pattern also found in cases of CR in other lan-guages. Both the long distance agreement and the appearance of the pronominal copyin a range of different grammatical functions14 would follow from the analysis of CR

proposed by Asudeh and Toivonen (2012) which involves anaphoric binding betweenthe matrix SUBJ (as the antecedent) and the embedded GF.

13It appears so far that this ‘object raising’ pattern is heavily restricted: the examples we give allinvolve embedded EOPVs.

14In MT the copy can be an OBJ, OBJθ , POSS or OBL OBJ.

(66) Donn-kom/hathink-2PL.ACC/3SGF.ACC

gèogob-t-u-hapleased.PV-2-PL-3SGF.ACC

She seems to have been pleased by you.You seem to have pleased her. MT

(67) Qis-u/hommeasure-3SGM.ACC/3PL.ACC

gèelib-homovercome.PV.3SGM-3PL.ACC

in-ngèas.SGM

DEF-sleepiness.SGM

They seem to have been overcome by tiredness.Tiredness seems to have overcome them. MT

(68) Qis-hameasure-3SGF.ACC

//

donn-hathink-3SGF.ACC

gaalready

bde-wstarted.PV.3-PL

j-i-stkerrh-u-ha3-FRM.VWL-disgust.IMP-PL-3SGF.ACC

She seems/appears as though they already started disgusting her. MT

If the observations in this section are on the right track, a promising approachwould be to extend the style of analysis outlined in Asudeh (2012) and Asudeh andToivonen (2012) to these data from the Arabic dialects. Such an analysis would needto take account of a number of syntactic differences between these constructions andEnglish Copy Raising, not least the lack of an intervening as if, like predication,which plays a crucial role in the syntactic part of that analysis. Since our concernhere is mainly with establishing and discussing the data patterns which we have in-vestigated, we leave these details of analysis for future work.

5 ConclusionThis paper has aimed to make a contribution to the syntactic description of the con-temporary Arabic vernaculars from the perspective of LFG. We have investigated twoclasses of verbs. We have shown that the dialects in question have a class of psychpredicates in which the experiencer is realized as a surface OBJ, and we have shownthat this argument shares the properties we associate with normal transitive objects.We then showed that the dialects have a class of aspectual verbs which should betreated as subject-subject raising verbs. The final section explored the interaction ofthese two classes of verbs, where we see a construction which bears a number ofresemblances to Copy Raising.

ReferencesAbdel-Massih, Ernest. 1979/2011. A Comprehensive Study of Egyptian Arabic, vol. 3.

University of Michigan.Al-Aqarbeh, Rania. 2011. Finiteness in Jordanian Arabic: A Semantic and Mor-

phosyntactic Approach. Ph.D. thesis, University of Kansas.

Arka, I Wayan. 2000. Control theory and argument structure: Explaining controlinto Subject in Indonesian. Paper presented at the 4th International Malay andIndonesian Symposium.

Asudeh, Ash. 2005. Control and semantic resource sensitivity. Journal of Linguistics41:465–511.

Asudeh, Ash. 2012. The logic of pronominal resumption. Oxford: Oxford UniversityPress.

Asudeh, Ash and Ida Toivonen. 2012. Copy raising and Perception. Natural Lan-guage and Linguistic Theory 30/2:321–380.

Belletti, Adriana and Luigi Rizzi. 1988. Psych-verbs and theta-theory. Natural Lan-guage & Linguistic Theory 6:291–352.

Camilleri, Maris, Shaimaa ElSadek, and Louisa Sadler. 2013. A cross dialectal viewof the Arabic Ditransitive Alternation. submitted to Acta Hungarica Linguistica.

Fassi Fehri, Abdelkader. 2012. Key Features and Parameters in Arabic Grammar.Amsterdam: John Benjamins.

Firanescu, Daniela Rodica. 2010. Do you still love Feiruz? The modal b@’ in spokenArabic from Syria. Monde Arabe 7:123–142.

Haddad, Youssef A. 2012. Raising in Standard Arabic: Backward, Forward, andNone. In R. Bassiouney and E. G. Katz, eds., Arabic Language and Linguistics,pages 61–78. Georgetown: Georgetown University Press.

Holes, Clive. 2004. Modern Arabic: Structures, Functions and Varieties: RevisedEdition. Washington,DC: Georgetown University Press.

Kim, Young-Yoo and Richard Larson. 1989. Scope interaction and the syntax ofpsych verbs. Linguistic Inquiry 20:681–688.

Landau, Idan. 2010. The Locative Syntax of Experiencers. Boston, Mass: MIT Press.Landau, Idan. 2011. Predication vs aboutness in copy raising. Natural Language &

Linguistic Theory 29(3):779–813.Peterson, John. 2009. ”Pseudo-verbs”: An analysis of non-verbal (co-)predication in

Maltese. In B. Comrie, R. Fabri, E. Hume, M. Mifsud, T. Stolz, and M. Vanhove,eds., Introducing Maltese Linguistics, pages 181–205. Berlin: John Benjamins.

Sadler, Louisa. 2013. Psych Predicates and an unexpected case of copy raising?Paper presented at CSSP 2013.

Sadler, Louisa and Maris Camilleri. 2013. Ditransitive Predicates and Dative Argu-ments in Maltese. Lingua pages 36–61.

Sells, Peter. 2007. Finiteness in non-transformational syntactic frameworks. InI. Nikolaeva, ed., Finiteness: Theoretical and Empirical Foundations, pages 59–88. Oxford: Oxford University Press.